1 of 51

Introduction to

CCP-EM

Colin Palmer

S2C2 CCP-EM workshop

10 Nov 2020

2 of 51

Collaborative Computational Project for Electron cryo-Microscopy

Support users and developers in computational aspects of biological EM

Hosted by STFC Rutherford Appleton Laboratory

Alongside CCP4 core team – shared expertise between projects

What is CCP-EM?

CCP-EM & CCP4 | RCaH

eBIC | DLS

Tom

Burnley

Martyn

Winn

Colin

Palmer

Agnel

Joseph

Jola

Mirecka

Matt Iadanza

3 of 51

Collaborative Computational Project for cryo-EM

Thanks to the many people helping to teach this workshop!

4 of 51

CCP-EM training workshops

Previously…

5 of 51

CCP-EM training workshops

Now

6 of 51

CCP-EM training workshops

Now

Our first virtual workshop

Please be patient!

At our in-person workshops we normally teach about 20 people. Now we have 250…

Great to reach so many people, but we don’t have capacity to give individual attention

7 of 51

Setup for tutorial participants

Software

Should all be installed now

If problems, raise in Q&A during tutorials

Data

Please download from https://www.ccpem.ac.uk/training/s2c2_workshop_2020/s2c2_workshop_2020.php

8 of 51

Suite of utilities for EM data processing

Common Python framework

Uses some CCP4 programs

Download from ccpem.ac.uk

Linux & Mac

Free for academic use, fee for commercial

Bugs & requests:

ccpem@stfc.ac.uk

CCP-EM software suite

9 of 51

CCP-EM website

Documentation

Tutorials and lectures

CCP-EM mailing list

Support for CCP-EM and RELION

https://www.jiscmail.ac.uk/CCPEM

CCP-EM software – more information

10 of 51

CCP-EM Spring Symposium

Annual UK cryo-EM conference

Talks on all aspects of cryo-EM including software

Recordings on YouTube (search “CCP-EM”)

Proceedings in Acta Cryst D

CCP-EM software – more information

11 of 51

SBGrid webinars

See SBGrid website or YouTube

CCP-EM papers

Collaborative Computational Project for Electron cryo-Microscopy. Acta Cryst. D71:123–126 (2015)

Recent developments in the CCP-EM software suite. Acta Cryst. D73:469–477 (2017)

CCP-EM software – more information

12 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

13 of 51

This workshop

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

14 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

15 of 51

State-of-the-art software for Single Particle Reconstruction

CCP-EM includes pre-compiled RELION binaries for Linux and Mac

CUDA GPU support on Linux

v3.1 in CCP-EM 1.5

Plan to integrate RELION more closely with the rest of the suite in future

RELION

Sjors Scheres

16 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

17 of 51

Maps from cryo-EM look like maps from X-ray crystallography...

EM maps are different

EM 3.3Å

MX 3.4Å

20S proteasome

Li et al, Nature Methods, 2013

18 of 51

Maps from cryo-EM look like maps from X-ray crystallography...

...but some important differences:

  • Map is phased!
    • Better quality initially
    • Doesn’t change during model building
  • Densities represent Coulomb potential, not electron density
  • Box size is arbitrary
  • Local resolution can be highly variable
  • Map sharpening affects refinement

Crystallography software (e.g. CCP4) can often handle cryo-EM maps, but needs care

EM maps are different

19 of 51

MRC to MTZ

Global map sharpening / blurring

Try an array of sharpening factors

Map sharpening

20 of 51

MRC to MTZ

Global map sharpening / blurring

Try an array of sharpening factors

Map sharpening

Garib Murshudov

21 of 51

MRC to MTZ

Global map sharpening / blurring

Try an array of sharpening factors

Visualise multiple sharpened / blurred maps in Coot

Expect local variation...

Map sharpening

Rob Nicholls

22 of 51

Locally adaptive map sharpening

Fits experimental map to local amplitude profile from atomic model B-factors

Requires a refined model (for now!)

Iterative process of model building and map improvement

LocScale

Arjen

Jakobi

Jakobi et al. eLife (2017) 6:e27131

23 of 51

LocScale

Arjen

Jakobi

Jakobi et al. eLife (2017) 6:e27131

Locally adaptive map sharpening

Fits experimental map to local amplitude profile from atomic model B-factors

Requires a refined model (for now!)

Iterative process of model building and map improvement

24 of 51

Local Agreement Filter for Transmission EM Reconstructions

Compares band-passed half maps to identify locally-shared features

Preserves shared signal, suppresses noise

LAFTER

Ramlaul, Palmer & Aylett (2019) J. Struct. Biol 205:30–40

Chris Aylett

25 of 51

Local Agreement Filter for Transmission EM Reconstructions

Compares band-passed half maps to identify locally-shared features

Preserves shared signal, suppresses noise

High contour: strong features remain similar

Low contour: weak noise features are removed

LAFTER

Ramlaul, Palmer & Aylett (2019) J. Struct. Biol 205:30–40

Chris Aylett

High contour

Low contour

Original EMD-2847

LAFTER filtered

26 of 51

Applies multiple hypothesis testing to cryo-EM maps

p-values adjusted for control of False Discovery Rate

Voxel values give a measure of confidence that we can discriminate signal from noise

At a threshold of 0.99 (1% FDR), at least 99% of the voxels truly indicate positive density signal in the map

Confidence Maps

Beckers, Jakobi & Sachse (2019)

Max Beckers

27 of 51

Paper in Acta D

Practical guidance for use and interpretation

doi:10.1107/S2059798320002995

Confidence Maps

28 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

29 of 51

Rigid-body fitting

Alexei

Vagin

MOLREP

Fast docking of molecular models

High resolution EM maps

Find multiple copies

Reference to target sequence correction

30 of 51

Rigid-body fitting

Alan Roseman

Dock-EM

Docking atomic models at medium to low resolution

Exhaustive 6D rigid body search

Target region of interest

Solutions ranked by cross-correlation coefficient (CCC)

View best hits in Chimera

31 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

32 of 51

Automated model building in high-resolution maps (<4Å)

Buccaneer: amino acid

Nautilus: nucleic acid

Requires map and sequence

Buccaneer & Nautilus

Kevin Cowtan

Scott Hoh

Find Cα seed positions

Grow chain fragments

Join overlapping fragments

Link adjacent fragments, assign & correct sequence, filter poor-quality fragments

Prune inconsistent fragments

Rebuild side chains

33 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

34 of 51

Flexible fitting

Agnel Joseph

Flex-EM

Model refinement at medium resolutions

Flexible fitting of rigid body domains to EM maps

Real space MD refinement

Rigid bodies detected based on clusters of secondary structure elements

Maya Topf

35 of 51

REFMAC

Garib Murshudov

Model refinement at high resolution (<~5Å)

Aims:

  • Model should agree with the observed data
  • Model must be chemically and structurally sensible

Automatic handling of EM maps:

  • Map to MTZ conversion
  • Electron structure factors
  • Map sharpening

Global or local refinement modes

Additional restraints

Oleg Kovalevskiy

Rob Nicholls

+

Data

Atomic model

Fit and refine

36 of 51

REFMAC

More information:

37 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

38 of 51

CCP-EM Coot built from “refinement” branch – v0.9.2

Now on both Mac and Linux

New features for cryo-EM

Coot

Paul Emsley

39 of 51

CCP-EM workflow

Single Particle Reconstruction

Map Optimisation

Docking /�Model Building

Automated Refinement

Buccaneer

Flex-EM

Molrep

Coot

Chimera

MRC to MTZ

LocScale

Dock-EM

RELION

Confidence Maps

Interactive Refinement

LAFTER

Refmac

Validation

Model Validation

40 of 51

What leads to overfitting?

  • Insufficient data (low resolution, partial occupancy)
  • Ignoring data (cutting by resolution)
  • Sub-optimal parameterisation
  • Bad weighting
  • Excess of imagination!

Need for validation

Garib Murshudov

41 of 51

Another observation common to almost all the deposited models based on high-resolution maps is that they seem to lack the final quality control. The presence of very doubtful multiple conformations of the side chains, poor geometry of the model in comparatively clear regions of the maps, location of the side chains outside of the clear density, or the occurrence of interatomic clashes may indicate the difficulty of manual inspection of these very large structures....

Nevertheless, more attention needs to be paid to such problems that are not easily solved by purely automated means.

Need for validation

Wlodawer et al., (2017) Structure 25:1589

Validation for cryo-EM is still immature and developing rapidly

We need better metrics and better education – but the situation is improving

Beware of errors in models from public databases!

Check important parts yourself

Please deposit your data!

42 of 51

Automated REFMAC half map validation pipeline

Checks for over-fitting in refinement

Requires 3 input maps:

  • Final map
  • Half map 1
  • Half map 2

Refinement protocol performed twice:

  • vs full map
  • vs half map 1

Half-map Validation

43 of 51

Half-map Validation

FSCwork: model refined against half map 1; compared to half map 1

FSCfree: model refined against half map 1; compared to half map 2

Not over-fitted

Over-fitted

44 of 51

CCP-EM task to run multiple validation metrics:

  • Geometry (bond/angle/dihedral): Molprobity
  • Cα geometry / peptides: CaBLAM
  • Local density fit: TEMPy SMOC
  • Global density fit: REFMAC FSC

To-do list of residues to check

Problems grouped in clusters

Model Validation

Agnel Joseph

45 of 51

WT Validation Symposium: 18 – 20th Nov

  • Cryo-EM Validation in the Age of SARS-CoV-2: Methods, Tools and Applications”

  • Virtual zoom meeting (free)

  • Details and registration at http://bit.ly/valid-sars

  • Topics include:
    • Map validation
    • Model validation
    • Map to model validation

46 of 51

Other recent additions to CCP-EM

cryoEF v1.1

Quantify spread of particle angle distribution, recommends tilt angles to minimize bias

Naydenova K & Russo CJ. Nat Commun (2017)

SIDESPLITTER

Binary and wrapper to work with RELION 3.1,

Local filtering approach to mitigate local overfitting

Ramlaul K et al. J Struct Biol (2020)

Haruspex

Deep learning approach to identify secondary structures in maps

Thorn A et al. Angewandte Chemie (2020)

Difference maps

(TEMPy and LocScale)

Global/local scaling based map-model difference

Joseph AP et al. JCIM (2019)

EMDA

Python library for interpretation of multiple maps and models, Local correlation and difference calculation

Rangana W and Garib M, MRC-LMB

Recent developments

47 of 51

SIDESPLITTER

Modification of LAFTER algorithm to maintain strict half-set separation

Reduces local over-fitting in refinement

Can help a lot when local resolution is highly variable

See Chris Aylett’s talk from the Spring Symposium

Published recently: doi:10.1016/j.jsb.2020.107545

Ramlaul, Palmer & Aylett (2019) bioRxiv

48 of 51

Difference Maps

Difference calculation with amplitude scaling

Published recently: doi:10.1021/acs.jcim.9b01103

Joseph et al. (2020) JCIM

49 of 51

Haruspex

Annotation of secondary structures in maps by deep learning

See Andrea Thorn’s lecture

Published recently: doi:10.1002/anie.202000421

Mostosi et al. (2020) Angew. Chem. Int. Ed. 59:2–10

50 of 51

CCP-EM core team

  • Tom Burnley
  • Colin Palmer
  • Agnel Praveen Joseph
  • Jola Mirecka
  • Matt Iadanza
  • Martyn Winn

CCP4 core team

STFC SCD

  • Alan Kyffin

DLS / eBIC staff

Birkbeck

  • Maya Topf

Acknowledgements

Imperial College London

  • Chris Aylett

Francis Crick Institute

  • Peter Rosenthal

EBI

  • Gerard Kleywegt
  • Ardan Patwardhan

MRC-LMB

  • Garib Murshudov
  • Sjors Scheres
  • Paul Emsley
  • Rob Nicholls
  • Oleg Kovalevskiy

University of York

  • Kevin Cowtan
  • Soon Wen ‘Scott’ Hoh
  • Jon Agirre

University of Manchester

  • Alan Roseman

TU Delft

  • Arjen Jakobi

EMBL / FZ Jülich

  • Max Beckers
  • Carsten Sachse

University of Würzburg

  • Andrea Thorn

51 of 51

Setup for tutorial participants

Reminder

Please download the tutorial data from https://www.ccpem.ac.uk/training/s2c2_workshop_2020/s2c2_workshop_2020.php